Attribution and prediction of maximum temperature extremes in SE Australia
- Publication Type:
- Journal Article
- Citation:
- Procedia Computer Science, 2014, 36 (C), pp. 612 - 617
- Issue Date:
- 2014-01-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
© 2014 Published by Elsevier B.V. Over half of Australia's population occupy its southeastern quadrant. Temperature records for the 56-year period 1958-2013 reveal increasingly hot summers since the 1990s, with daily maximum temperatures reaching 10 °C above normal. The change in monthly mean maximum temperatures (∼1 °C to 1.5 °C above the long term mean) far exceeds the natural variability expected over a half-century. Numerous maximum temperature records have been set and the extreme heat poses a major socioeconomic threat. This work seeks climate drivers that are useful predictors of the warm mean monthly values of maximum daily temperatures for January, in southeastern Australia. The data for January 1958-2013 from one representative site, Tibooburra, is coded, in a binary sense (excessive heat-yes/no), and for actual temperature anomalies. One challenge in analyzing these data is the short records relative to the numerous possible climate drivers of excessive heat. The variables are a combination of ocean and atmospheric climate drivers plus their high and low frequency filtered values from wavelet analysis. Several feature selection methods are applied to produce a compact set of predictors exhibiting good generalization properties. Results of cross-validation of logistic regression, with and without threshold adjustment, show that cold air blocking, and teleconnection patterns, such as the Southern Annular Mode (SAM), have statistical skill (best classification Heidke skill score = 0.34) in forecasting extreme heat for binary forecasts, with correct forecasts exceeding 75% of cases. For predicting actual monthly anomalies, support vector regression and bagged trees explain anomaly temperatures with mean absolute error of 1.4 °C and 1.3 °C.
Please use this identifier to cite or link to this item: